SharpNEAT GUI / Main Control Window

 

 

Progress Monitoring Graphs

The graphs on this screen are drawn using the excellent free graphing control ZedGraph. Thanks to Mike Woodhouse for adding these graphs to SharpNEAT.

 

 

Species List / Visualization

 

 

Function Regression Experiment View

Here we are regressing a function of the form y=f(x), where f(x) is made up from joining of a number of simple mathematical functions - simple curves and straight lines. The resulting function shown here has many sharp edges to it and is being regressed by NEAT using the standard sigmoid activation function, which is a smooth continuous function. This particular experiment then demonstrates the difficulty of regressing a function with sharp edges using a continuous activation function. This is a good argument for investigating alternative activation functions and/or using combinations of activations functions. In fact Mattias Fagerlund performed some successfull experiments in this area using his own DelphiNEAT.

 

 

SimpleOCR Experiment View

 

 

Pole Balancing Simulator EXE

Here we see the tricky problem of double pole balancing and the even trickier problem of double pole balancing without a velocity input. Two poles attached to a cart can in fact be balanced if they are of different lengths. The lengths shown here are actually not representative - the short pole is actually much shorter but is shown longer to allow us to see its angle more clearly.

 

 

Prey Capture Simulator EXE

Here you can see the prey (in green) and our agent (red) with its 8 radial sensor areas shown. The agent cannot see beyond the sensor ranges shown and thus must employ a search strategy before successfully using its sensor inputs to finally home in on the prey.

 

 

Tic Tac Toe Game / Simulator EXE